EDC-based transceiver standard wrestles with distance issues
Today’s enterprise networks have a large installed base of legacy 62.5-µm multimode fiber (MMF) with link lengths up to 300 m that currently support 1-Gbit/sec transmission. Increased data traffic is driving the need to upgrade those links to 10 Gbits/sec, but at that rate, MMF is subject to modal dispersion that can limit transmission distance. Advances in silicon have enabled the development of electronic dispersion compensation (EDC) ICs, which many believe will become a key building block of all 10-Gbit/sec networks.
In November 2003, the IEEE established the 802.3aq Task Force to develop a cost-effective mechanism for upgrading 1-Gigabit Ethernet (GbE) MMF links to 10-GbE. The operative phrase, according to the task force, is “cost-effective.” There is already a standardized solution for 10-GbE upgrades, known as 10GBase-LX4, and first-generation LX4 products are already available from several transceiver manufacturers. The LX4 is a WDM-based approach, however, requiring four sets of lasers, four receivers, and the associated electronics, which may make its widespread deployment cost-prohibitive.
The task force is currently working on a 10-Gbit/sec serial alternative to the LX4 parallel solution, dubbed the 10GBase-LRM. The serial device would include just one laser, one receiver, and a set of relatively simple electronics-a laser driver, a transimpendence amplifier, and an EDC chip-making it more easily manufactured and therefore less expensive. But the LX4 has one key advantage: It supports distances up to 300 m as specified by the riser cabling standards. As the task force prepared for its most recent meeting in late September, the LRM was only targeting 220 m of MMF.
The 10GBase-LRM standard will specify support for multiple fiber types, including both 62.5- and 50-µm MMF. While only about 5% of the installed base, 160/500-MHz•km MMF-the 62.5-µm FDDI-grade fiber installed in the early 1990s-is providing the biggest challenge for the IEEE committee. FDDI-grade fiber has the lowest-level bandwidth and is therefore the most difficult to support; moreover, its characteristics are not very well known. “We’re spending a lot of time working on models for what this type of fiber actually looks like,” explains David Cunningham, chair of the IEEE 802.3aq Task Force and senior manager of strategic R&D and technology in Agilent Technologies’ Fiber Optic Product Division (Palo Alto, CA). “It’s important that we understand that in terms of being able to say what distance we can support on the fiber.”
Right now, that distance is 220 m, which is the minimum distance supported by the 1-GbE standard the LRM is designed to replace. “It was felt that 220 m was the shortest distance that would make sense for the installed base,” says Cunningham. “If we could tradeoff complexity and cost so that eventual products based on the standard would be very cost-effective, as long as they met at least [220 m], they would still make reasonable sense. But if they fell below it, then they would make no sense at all.”
The distance limitations of the 10GBase-LRM standard have become a minor point of contention; there are differences of opinion even within the IEEE committee. “There’s been a big push from the systems’ vendors to say that it would be very desirable if not mandatory to reach 300 m consistent with the building cabling standards,” reports Bob Zona, director of marketing for the optical-module group at Intel (Santa Clara, CA). It comes down to “a tradeoff between the distance in that 220-300-m range, the cost of the solution that results, and the percentage of coverage of all the fibers that exist,” he says. While Intel “hasn’t taken a firm position yet,” Zona admits, “there are strong reasons to try to achieve 300 m.”
All the EDC vendors interviewed here say they have designed or are designing their devices to support the longer distance, regardless of the standard. “We expect the EDC-based solution will be just like the 1-GbE solution,” reports Abhijit Phanse, president and CEO of Scintera Networks (San Jose, CA). “Although it’s specified for 220 m, in reality, it will probably work well beyond 300 m and meet the requirements of a very, very large percentage of the installed base. What really matters is what percentage of the installed base do you actually cover, and the EDC solution will cover in the 90s.”
Achieving the longer distance is not an insurmountable technological hurdle; it’s achieving the distance while also meeting the stringent requirements of small-form-factor 10-Gbit/sec transceivers. “With a high-power engine, you can do even 300 m over multimode fiber without a problem, but you have to keep the costs down to fit into XFP and XENPAK modules,” explains Oswin Schreiber, senior product marketing manager at AMCC (San Diego). “You are limited power-wise.”
The ultimate goal of the IEEE 802.3aq Task Force is “to find the economic sweet spot,” adds Michael Lawton, strategic marketing program manager at Agilent’s Fiber Optic Product Division. “If we add too much complexity, we end up with a product that is essentially similar to the LX4 and isn’t going to enable the market we want to create.”
According to Nick Weiner, chief technology officer at Phyworks (Bristol, UK) and editor of the 10GBase-LRM standard, that sweet spot needs to be somewhere in the neighborhood of $200 by 2006. Moreover, he says, “We believe the chip content can’t be any more than 20% of that, so we’re talking $40 for all of the chips inside the XFP.”
At press time, Cunningham was preparing for the task force’s interim meeting in September, during which the committee hoped to gain a better understanding of the bandwidth performance of the installed base of MMF. Fiber manufacturers and experts from the likes of Cambridge University were scheduled to present cable models and statistical plots for accurately evaluating the optical characteristics of the installed fiber.
The one overriding fact upon which everyone interviewed here agrees is that EDC is the critical technology that will enable the widespread deployment of 10-Gbit/sec links. “We see EDC not as another chip people need to buy but as a new function that needs to be integrated into existing chips once you get to 10 Gbits/sec,” says Phyworks EDC product manager Chet Babla.
Initially, IEEE 802.3aq heard three proposals, including one that specified multilevel encoding and another that required a special launch that would make the fiber appear as though it were singlemode. Both proposals were ultimately rejected in favor of the EDC proposal, which was by far the simplest of the three, says Cunningham. “EDC is perceived as a ‘cure-all’ for all kinds of high-speed networks,” notes Phanse. “There is just tremendous unanimous industry support. You’re going to see a very quick shift to a serial solution.”
Not everyone is quite so bullish, however. “As things settle down, if the standard does specify something that customers really want and the accompanying pieces are there, then we see it as a viable option,” admits Wiren Perera, marketing vice president at Mysticom (Netanya, Israel). “Here and now today, the solution that works over 300 m of multimode fiber is LX4-which is without EDC.” The company is nevertheless in the process of developing a 10GBase-LRM device to compete in a market that already includes Scintera, Phyworks, AMCC, Quake Technologies (Kanata, Ontario), and startup ClariPHY Communications (Newport Beach, CA).
The EDC vendors expect to see the first EDC-based LRM devices ahead of the standard’s expected late 2005/early 2006 ratification date. “It will be important to see what people like AMCC and others can achieve,” notes AMCC’s Schreiber. “People will definitely start using this technology before the standard is closed. There is a real need right now to go for longer distances at a lower cost point.”
EDC technology may be in its infancy, says Cunningham, but in “a few years’ time, it will probably become part of most optics at 10 Gbits/sec, and that’s why we’re backing it. We can see that it’s going to become a relatively generic building block, and it does certainly increase the performance of what you can achieve.”
Meghan Fuller is the news editor at Lightwave.